OpenAI submitted a confidential draft registration statement to the Securities and Exchange Commission on 8 June 2026. It has not filed a public S-1, and a direct search of SEC EDGAR on 21 August 2026 returns no OpenAI operating company as a registrant — only two dozen special purpose vehicles and secondary funds with “OpenAI” in their names. The September 2026 listing target that circulated in the summer has been superseded: chief financial officer Sarah Friar told staff on 19 August 2026 that the company will be public in 2027, or sooner if revenue inflects. A roughly 7 billion dollar employee tender offer closed on 10 August 2026 at a valuation of 852 billion dollars, unchanged since March.

Why it matters

Nothing here changes what an API call costs you tomorrow. But a company that is preparing an eventual prospectus is a company that will soon have to publish audited unit economics, and that is the first time anyone outside OpenAI will see what serving a token actually costs. Buyers negotiating multi-year commitments should treat the current pricing environment as pre-disclosure pricing, set by private capital rather than by public scrutiny.

The filing is confidential, and EDGAR shows nothing

The distinction matters more than most coverage has allowed.

A confidential draft registration statement is a submission, not a filing. Under the Jumpstart Our Business Startups Act, an issuer can hand the SEC a draft Form S-1 for review without publishing it. Nothing appears on EDGAR. No financial statements become public. The company retains the option to abandon the process entirely, and many do.

OpenAI announced the submission itself on 8 June 2026, saying it expected the news to leak and framing the step as optionality rather than commitment. Its own language was explicit that timing had not been decided.

We checked EDGAR directly. The company search for “openai” returns 24 entities: OpenAI Startup Fund I and its six SPVs, and a long list of series LLCs run by third parties to pool secondary exposure. None is the operating company. A full-text search across S-1 filings for OpenAI returns only a CoreWeave exhibit from March 2025.

So the accurate statement is this: OpenAI has begun the process that can lead to an initial public offering. It has not filed a prospectus. Anyone reporting that an S-1 has landed on EDGAR is wrong, and the error is checkable in about ninety seconds.

Anthropic did the same thing a week earlier, submitting its own confidential draft on 1 June 2026.

The revenue figure in circulation is out of date

The “roughly two billion dollars per month” number is real but stale. It described an annualised run rate near 25 billion dollars that held through the spring of 2026.

Bloomberg reported on 13 August 2026 that OpenAI’s annualised revenue run rate has passed 40 billion dollars, roughly double its pace in late 2025, with president Greg Brockman telling staff the monthly run rate grew more than 20 percent in July alone. That works out to something closer to 3.3 billion dollars per month.

Friar separately told investors around 14 August 2026 that enterprise revenue has overtaken consumer revenue, a crossover the company had previously guided to arrive at the end of the year.

None of these figures is audited. They are company statements relayed by reporters, and run rate is an annualisation of a single recent period, not a trailing twelve-month result. A prospectus would replace all of it with GAAP revenue. That has not happened.

The loss figure is a projection, not a disclosure

The 14 billion dollar loss for 2026 comes from The Information, which reported in February 2026 that it had seen internal OpenAI projections. The same reporting described cumulative losses of roughly 44 billion dollars across 2023 to 2028, before a swing to profit in 2029.

Three cautions. It is a forecast, not a result. It appears to describe a non-GAAP operating measure, and several analysts reading the same reporting have put the GAAP net loss materially higher. And it was made before the revenue acceleration Bloomberg described in August, which cuts in the company’s favour. OpenAI has not confirmed the number.

The tender offer held the valuation flat, and that is the signal

On 10 August 2026, OpenAI completed a buyback of roughly 7 billion dollars of employee shares at a valuation of 852 billion dollars, first reported by Bloomberg. The company funded the purchase itself rather than routing it to outside investors, as it had in a 6.6 billion dollar tender at 500 billion dollars in October 2025.

Two things follow. First, 852 billion dollars is the same mark set in the March 2026 round. The private valuation has not moved in five months, even as the revenue run rate roughly doubled. Second, a tender offer relieves the pressure that usually forces a listing. Employees who wanted liquidity have it. That is consistent with Friar’s 19 August guidance to 2027, and it is the most straightforward reading of why the September date went away.

Sam Altman has reportedly pushed advisers toward a trillion dollar listing valuation. That is an aspiration reported second-hand, not a price.

Anthropic reached the operating line first

The contrast is the story. Anthropic reported preliminary second-quarter 2026 revenue above 11.5 billion dollars and positive adjusted operating income — its first — according to reporting on 15 August 2026. The company describes the numbers as preliminary and the profit measure is adjusted, which does the usual work of excluding items.

Set against OpenAI’s position, the two labs are running different experiments. One has reached an operating profit on an adjusted basis. The other is scaling faster in absolute users and consumer reach, spending far more to do it, and heading toward public markets with a loss to explain.

Neither has published audited accounts. Both have submitted confidential drafts. Within roughly a year, both may have to show the same reader the same kind of statement, and that will be the first apples-to-apples comparison the industry has ever had.

What this means if you buy AI products

Prices are still being set by companies that answer to private investors, and one of them has now told staff it expects public-market discipline in 2027.

That is worth holding in mind alongside the price movement of the past two months. OpenAI cut GPT-5.6 Luna pricing by 80 percent on 6 August 2026, and the broader effect has been to move the competition to cost per useful answer rather than headline capability. The Luna cut sits at one end of a widening range; GPT-5.6 Sol sits at the other.

Assume current cheap tiers are promotional until proven otherwise. Read the pricing pages for expiry dates on introductory rates and for the threshold where long-context surcharges begin.

Prefer commitments with published rate schedules and exit terms over discounts that depend on a vendor’s continued willingness to lose money. Keep a tested fallback model for anything running in production — the cost of switching is the only real leverage you have.

Treat the eventual prospectus as a research document, not a market event. When it comes, it will contain gross margin by segment and compute commitments — the two numbers that actually determine what you will pay in 2028.

This is reporting on a company’s financial position. It is not investment advice, and shares in OpenAI are not available to the public.

What would change our reading

  • A public S-1 appearing on EDGAR under an OpenAI operating entity. That single event would change this story from preparation to fact.
  • A confirmed underwriting syndicate and exchange. Goldman Sachs and Morgan Stanley have been reported as leads, none of it confirmed by the company.
  • Any audited figure that contradicts the reported run rate.
  • A withdrawal, or a further slip past 2027. Confidential submissions expire quietly and often.
  • Anthropic’s audited operating margin. The adjusted profit claim is the load-bearing element of the contrast drawn above.

Sources